I view it as the study went looking for X, and to it's surprise didn't see X, but happened to see statistical significance in Y. However the study did not have the scope to determine why Y occurred.
I think the most likely thing that changed over time is Z: the first wave(s) of feminism began to liberate females from the traditional limitations of roles in society. However the true from of feminism is a recognition and reconstruction based around /all/ gender roles being bad for mental health, emotional health, and positive relations among individuals.
Men haven't had that liberation; a large part of society still believes in specific roles and responsibilities for men that coupled females are (at their option) exempted from. This probably leads to increased stress to both individuals and also an increased risk of negative feedback loops for the situation and everyone involved.
The inherent bias in society is even reflected in the premise of the study. It's tracking a specific type of male/female relationship and a specific correlation of role towards problems. Instead a more balanced study would look at the starting condition, determine what percentage of the population (restricting to males and females alone is OK, but including all sets of couples is better) fits different configurations, and tracking for outcomes as well as associated economic, emotional, mental, and physical health in the related intervals would be a better study design. The study might also have a follow-up for 'marriage failure' cases where isolation of the root causes is performed.